Oddbean new post about login | logout NEW - OpenAI, Anthropic, and Meta's AI models chose violence and escalated to nuclear strikes in simulated wargames, even in neutral scenarios. https://www.disclose.tv/id/ix9yjpmll1/ @disclosetv
nostr:nevent1qqsdtuw3a8qxgsvstzsk85np9tswm5nt0fzzpvrq6tlc49a7p2ackjcpr3mhxue69uhkummnw3ezucnfw33k76twv4ezuum0vd5kzmqzypqweavm2cqfsyl2fscgrvvan4f97akm7fg6tgpcnd975f7r56uc7qcyqqqqqqgt706d0